Staging strided access to input buffer in a guardwithif generates a pointless if statement

[abadams]

This code:

#include "Halide.h"

using namespace Halide;

int main(int argc, char **argv) {
    Func f, g;
    Var x, y;

    ImageParam im(UInt(8), 1);

    g(x) = im(2 * x) + im(2 * x + 1);

    g.vectorize(x, 32, TailStrategy::GuardWithIf);
    im.in().compute_at(g, x).vectorize(_0, 64, TailStrategy::GuardWithIf);

    g.compile_jit();

    return 0;
}

Generates this IR, which includes an always-true if statement:

  let t114 = g.extent.0/32
  for (g.s0.x.x, 0, t114) {
   allocate im_im_global_wrapper$0[uint8 * 64]
   if ((g.s0.x.x*32) < g.extent.0) {
    produce im_im_global_wrapper$0 {
     im_im_global_wrapper$0[ramp(0, 1, 64)] = im[ramp((((g.s0.x.x*32) + g.min.0)*2) - im.min.0, 1, 64)]
    }
   }
   consume im_im_global_wrapper$0 {
    g[ramp(g.s0.x.x*32, 1, 32) aligned(32, 0)] = im_im_global_wrapper$0[ramp(0, 2, 32)] + im_im_global_wrapper$0[ramp(1, 2, 32)]
   }
   free im_im_global_wrapper$0
  }

[steven-johnson]

Is this if statement likely to be optimized away by LLVM (ie can it be proven always true or false)?

[abadams]

Sadly it is not optimized away